FPGA Design And Implementation Of Independent Component Analysis Algorithms

Blind source separation (BSS) currently has prominent applications in many fields suchas the biomedical engineering, speech recognication and radar communication system.Independent component analysis (ICA) is a method of finding linear nonorthogonalcoordinate systems to address BSS problems; the goal is to performance a linear transformthat makes the resulting variables as statistically independent from each other as possible.Infomax and FastICA are two typical ICA algorithms, and widely used to solve BSS problems.It’s difficult to realize strictly real-time processing by adopt ICA, as for two main factors.Firstly, for ICA itself, it’s offline, and needs a greater number of sampled data for everyprocessing. Secondly, ICA involves many complicated arithmetic operations, such asmultiplication-accumulator unit, matrix computation, iteration processing and so on, in otherwords, the complexity of ICA algorithms are quite high.Therefore, applying some transformations to ICA algorithms to change their structures,utilizing fully their parallel characteristics and use appropriate logic devices, all are quitesignificant and necessary to achieve real-time processing in engineering adopting ICA. FPGAis very suitable for distributed algorithms implementation, and widely applied to the highspeed or real-time signal processing. So far in VLSI chip design, it is often used as simulationand validation platform.This paper has discussed the issue about discribting FastICA and Infomax through HDLcodes. Importantly, quite many VLSI design methods are presented here, and adopted inimplementations of many arithmetic computations and FastICA system, in order to obtain abalance amoung power consumption, area and speed. Firstly, it illustrates ICA theory atdetails, especially FastICA and Infomax algorithm. And then presents quite lots of algorithmsfor arithmetic operations and its optimized design framework, respectively. Moreover,Synplify RTL circuits and simulation results are provided here for partial modules. Throughanalyzing algorithm structures of Infomax and FastICA, we have described Infomax andFastICA system-level architectures, further, simulink diagram and system generator structureare both presented here for Infomax. In addition, according to FastICA algorithm, itssystem-level architecture based on those arithmetic modules and matrix operation modulesdesigned is depicted here at length, which is able to separate M channels mixed signals.